Dev

Growth by day and python by night. I also like football, travelling, music and art. Check my github for my experiments in LLMs, Agents and Data

[Dev @HackrLife]

https://t.co/iJ4NLnctz9

[Dev @HackrLife]

Stack Overflow's 2026 Developer Survey: the top 5 things developers use AI or agents for right now (n = 12,547):

- Code in a familiar area: 69.3%
- Debugging or refactoring: 63.8%
- Straightforward technical questions: 59.4%
- Writing or improving tests: 58.1%
- Code in an unfamiliar area: 56.0%

The same survey: 48% trust AI when the answer is easy to validate.

If you build a dev tool or an internal agent, make every output checkable in one step:
1. Return a diff instead of a rewritten file
2. Run the tests and show the result next to the change
3. Link the doc or line of code each claim came from

https://t.co/cXpbj0UabH

[Dev @HackrLife]

For idea 2, if your site sits behind Cloudflare you can see a version of this today.

Cloudflare dashboard > your domain > AI Crawl Control:

- Crawlers tab: each AI crawler, its operator, request count, and an Allow or Block action
- Metrics tab: most popular paths, status codes, requests over time
- Filters by crawler, operator, hostname and path
- Referrals from AI tools (paid plans only)

[GREG ISENBERG @gregisenberg]

The MOST obvious (but REALLY good) STARTUP ideas right now:

1. Zestimate for company data (what your data is worth to an AI lab)

2. Google Analytics for agents (see which agents visit your site and what they want)

3. Y Combinator for 1 person companies (back solo founders running a team of agents)

4. Okta for agents (who each agent is and what it's allowed to touch)

5. Marie Kondo for your company's data (clean it up so agents can use it)

6. FICO for agents (a trust score before an agent can spend money)

7. Geico for agents (insurance for when an agent makes an expensive mistake)

8. Datadog for agents (see every step your agents took)

9. Twilio for agents (phone numbers and texting for agents)

10. Upwork for agents (hire an agent for a job instead of a freelancer)

11. ESPN for AI (every model launch covered like a big game).

12. Shutterstock for robot training videos (get paid to film yourself doing chores)

13. Credit Karma for what AI knows about you (see it, fix it, delete it)

14. Norton for AI scams (protect your parents from fake voices and emails)

15. PayPal Buyer Protection for agent purchases (fixes it when your agent buys the wrong thing)

16. Geek Squad for AI (set up personal agents for families and small businesses)

17. Dell for private AI (AI in a box for firms that won't use the cloud)

18. Carfax for used GPUs (a health report before you buy secondhand)

19. Airbnb for idle GPUs (rent out computing power you're not using)

20. Rotten Tomatoes for AI models (real reviews instead of benchmarks)

21. G2 for skills (find the best skills for your agent)

22. Dropbox for your AI memory (take what your agent knows about you to any agent)

23. Canva for physical products (design it with AI and get it made)

24. ADT for your personal agent (alerts when your agent does something it shouldn't)

25. Hertz for robots (rent one for a move, a deep clean or an event)

26. Harvard Business School for internet businesses (a real education in building online)

27. Workday for agents (onboard, manage and offboard your digital workers)

28. Getty Images for your face and voice (license your likeness to AI safely)

29. Meta Muse for older adults (and other high LTV niche)

30. Google Ads for agents (pay to be what the agent recommends)

31. Loom for agents (a walkthrough of everything your agent did)

32. Plaid for agents (approved access to banks, airlines and stores)

33. Jiffy Lube for vibe coded apps (quick fixes when your AI built app breaks)

34. Ancestry for datasets (where your training data came from)

35. The organic label for human-made content (proof a person made it)

36. Strava for builders (track what you ship every week)

37. Honey that haggles for you (your agent negotiates the price)

38. SEM Rush for personal agent platforms (like Muse etc)

39. Frequent flyer miles for agents (loyalty programs built for agents)

[Dev @HackrLife]

This rule loop maps well onto Claude Code's memory files if any of your bots run there.

From the Claude Code docs:
- One topic per file in .claude/rules/ (x-posts.md, newsletter.md). Add paths: frontmatter and a rule loads only when Claude works on matching files
- Keep each CLAUDE.md under 200 lines. Longer files use more context and reduce adherence

https://t.co/hbFGBgAjcU

[ericosiu @ericosiu]

How to make your Grok Bots stop repeating the same mistakes

Every correction becomes a rule the bots read before they work. 190 corrections in 3 weeks turned into 99 rules

The full loop is in the image.

How it works:

1. Log every correction the same day

> the exact words, quoted
> the asset it was about
> the rule it maps to

If a rule already covers it, mark it a repeat and make that rule more concrete

2. Give every rule an ID, a why and a tag

> the rule in one line
> the data or feedback behind it
> which bots it applies to: X, newsletter, LinkedIn, graphics

Every bot reads the shared rules plus its own tag before it starts work

3. Let results decide which rules stay

> X posts get checked at 24h and 72h against the 30-day median
> 2x the median, or bookmarks at 1% of impressions, is a win
> under 0.75x is a loss
> 3 straight losses in one format pauses it for 2 weeks

4. Run a daily check

Our Bot Improver checks the rules every day. The daily loop check flags anything slipping, like a weekly Article with no approved draft by Wednesday

5. Rewrite the rules every Monday

Our Chief of Staff bot fills a 30-minute review from the logs, then we confirm, add or retire rules. The number we watch is repeat feedback. The target is 0

Start with the last 3 things you corrected your AI on and turn each one into a rule

[Dev @HackrLife]

The 12-hour cache costs more to write, so check the break-even. From OpenAI's prompt caching docs (GPT-5.6+):

- Cache reads: 0.1x the normal input rate
- Cache writes: 1.25x on the default 30-minute TTL, 2.5x on 12h
- Turn it on with prompt_cache_options.ttl: "12h" (Responses API only)

If your calls reuse the same prefix more than 30 minutes apart, every call on the default TTL pays a fresh write. Cost in units of one uncached prefix:
2 calls: 2.5 (30m) vs 2.6 (12h)
3 calls: 3.75 vs 2.7
5 calls: 6.25 vs 2.9

So 12h pays from the third call. The prefix must be identical and 1,024+ tokens; usage.input_tokens_details.cached_tokens shows your hits.

[OpenAI Developers @OpenAIDevs]

You can now choose 12-hour prompt caching in the API.

Persistent agents, like dots, reuse context throughout the day.

You can keep shared instructions, tools, and context cached longer and pay the lower cached-input rate, even when requests are hours apart.

[Dev @HackrLife]

For the "fair test" step, Meta has one built in: an A/B test from Experiments in Ads Manager.

It randomly splits the audience so each person sees only one variation, which two separate ad sets with overlapping audiences don't give you. You test 2 to 5 variations, pick one key metric that decides the winner (cost per purchase here, not CTR), and Meta reports a confidence figure: the chance you'd get the same result if you ran it again.

Low confidence means keep spending or keep the current winner.

[Phil Kiel @PhilKiel]

When Meta puts 90% of spend into one ad, that’s usually the auction telling you something.

Meta looked at all 10 and predicted which one would convert best for the people it could reach. It’s often right, especially with only a few days of data.

Forcing every ad to 2,000 to 3,000 impressions has a few problems:

That’s nowhere near enough data. At a 1% CTR it’s 20 to 30 clicks per ad, so you’re reading noise.

CTR and CPC winners often aren’t CPA winners. Curiosity clicks are cheap. Buyers aren’t.

Every pause resets delivery, so the ad set never settles.

You end up spending budget on the ads Meta had already ruled out.

What I’d do instead on a small budget:

Test 3 to 5 ads per ad set. 10 is too many at $5k a month.

Make each ad a different concept. Small variations just compete with each other.

If you think an ad deserves more spend, put it in its own ad set with its own budget. That’s a fair test.

Give each test enough spend to get some purchases, then judge on CPA.

Leave the winner alone and add the next batch of concepts around it.

If the ad you rescued wins in its own ad set, Meta got it wrong. Most of the time it doesn’t.

[Dev @HackrLife]

A cheap way to build gate 2 without another model call: make the writer output every number with its source URL, fetch each page, and fail the draft if the exact figure isn't in the page text.

That catches invented and mistyped stats for the cost of a fetch. It won't catch a real number used out of context, so keep a human read for claims that carry the piece.

[Dev @HackrLife]

17/ The blueprint, with the rollout flow, build order and failure points: https://t.co/Xt7CXZXlIW

Source: Wharton Human-AI Research and GBK Collective, 2026 AI Adoption Report.

[Dev @HackrLife]

1/ 🧵 Wharton and GBK Collective surveyed about 1,400 people at large US companies on AI this July. 28% of managers report strongly positive ROI.

What that group did differently, in 10 steps over 17 posts, each with a number from the report.

[Dev @HackrLife]

16/ The method in one post: track up to three growth metrics, standardise on one platform, name an AI owner, redesign the org chart, start agents with sign-off, monitor them automatically, assign error ownership, train and ask frontline staff.

[Dev @HackrLife]

A concrete version of "check whether it succeeded": the Stop hook in Claude Code.

Add a Stop hook in settings.json that runs your test suite. If tests fail, the script prints {"decision": "block", "reason": "3 tests failing"} and Claude keeps working with that reason instead of ending its turn. Exit code 2 does the same.

So the agent can't call the job done until the check passes. Docs: https://t.co/zh8LPAAcZG